Assessing Diabetic Retinopathy Staging With AI: A Comparative Analysis Between Pseudocolor and LED Imaging

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Diabetic retinopathy (DR) is a leading cause of vision loss.
• Accurate staging of DR is crucial for timely treatment and management.
• Artificial intelligence (AI) systems offer potential for automated DR screening and staging.

Purpose of the Study:

• To compare the diagnostic performance of an AI-based DR staging system across different imaging modalities: pseudocolor, simulated white light (SWL), and light-emitting diode (LED) cameras.
• To evaluate the agreement between AI staging and human expert grading for DR severity.

Main Methods:

• A cross-sectional study involving 362 eyes from 189 diabetic patients.
• Imaging was performed using both Optos ultra-widefield pseudocolor and iCare DRSplus LED confocal cameras, with and without SWL.
• AI analysis (EyeArt v2.1) was compared against ground truth established by human graders using dilated fundus examinations.

Main Results:

• The LED camera showed higher sensitivity and specificity for proliferative DR and better agreement with ground truth (wκ = 0.71) compared to pseudocolor imaging.
• Adding SWL to pseudocolor imaging decreased diagnostic performance (wκ = 0.55).
• Peripheral lesions were found to significantly reduce the likelihood of correct DR staging.

Conclusions:

• LED cameras exhibit superior accuracy for identifying advanced diabetic retinopathy stages compared to pseudocolor cameras.
• AI systems require specific training for ultra-widefield imaging to account for peripheral lesions and ensure accurate DR staging.